algos.yaml

float:
  any:
    DolphinnPy:
      docker-tag: ann-benchmarks-dolphinn # Docker tag
      module: ann_benchmarks.algorithms.dolphinnpy # Python class
      constructor: DolphinnPy # Python class name
      run-groups:
        base:
          args: [[10, 50, 100, 200, 1000, 2000]]
    faiss-lsh:
      docker-tag: ann-benchmarks-faiss
      module: ann_benchmarks.algorithms.faiss
      constructor: FaissLSH
      run-groups:
        base:
          # When @args is a list, the result is the Cartesian product of all of
          # the things it contains; entries that aren't a list will be treated
          # as lists of length one.
          args: [[32, 64, 128, 256, 512, 1024, 2048, 4096]]
          # This run group will produce eight algorithm instances:
          # FaissLSH(32), FaissLSH(64), and so on up to FaissLSH(4096).
    faiss-ivf:
      docker-tag: ann-benchmarks-faiss
      module: ann_benchmarks.algorithms.faiss
      constructor: FaissIVF
      base-args: ["@metric"]
      run-groups:
        base:
          args: [[5, 10, 20, 50, 100, 200, 400, 800, 1000]]
          query-args: [[1, 2, 3, 4, 5, 8, 10, 20, 50, 100, 200]]
    flann:
      docker-tag: ann-benchmarks-flann
      module: ann_benchmarks.algorithms.flann
      constructor: FLANN
      base-args: ["@metric"]
      run-groups:
        flann:
          args: [[0.2, 0.5, 0.7, 0.8, 0.9, 0.95, 0.97]]
    panns:
      docker-tag: ann-benchmarks-panns
      module: ann_benchmarks.algorithms.panns
      constructor: PANNS
      base-args: ["@metric"]
      run-groups:
        five-trees:
          args: [5, 20]
        ten-trees:
          args: [10, [10, 50]]
        hundred-candidates:
          args: [[10, 20, 40], 100]
    annoy:
      docker-tag: ann-benchmarks-annoy
      module: ann_benchmarks.algorithms.annoy
      constructor: Annoy
      base-args: ["@metric"]
      run-groups:
        annoy:
          args: [[100, 200, 400]]
          query-args: [[100, 200, 400, 1000, 2000, 4000, 10000, 20000, 40000,
                      100000, 200000, 400000]]
          # This run group produces 3 algorithm instances -- Annoy("angular",
          # 100), Annoy("angular", 200), and Annoy("angular", 400) -- each of
          # which will be used to run 12 different queries.
    nearpy:
      docker-tag: ann-benchmarks-nearpy
      module: ann_benchmarks.algorithms.nearpy
      constructor: NearPy
      base-args: ["@metric"]
      run-groups:
        nearpy:
          args: [[10, 12, 14, 16], [5, 10, 20, 40]]
        extra:
          args: [16, [5, 10, 15, 20, 25, 30, 40]]
    bruteforce:
      docker-tag: ann-benchmarks-sklearn
      module: ann_benchmarks.algorithms.bruteforce
      constructor: BruteForce
      base-args: ["@metric"]
      run-groups:
        empty:
          args: []
    bruteforce-blas:
      docker-tag: ann-benchmarks-sklearn
      module: ann_benchmarks.algorithms.bruteforce
      constructor: BruteForceBLAS
      base-args: ["@metric"]
      run-groups:
        empty:
          args: []
    dummy-algo-st:
      docker-tag: ann-benchmarks-sklearn
      module: ann_benchmarks.algorithms.dummy_algo
      constructor: DummyAlgoSt
      base-args: ["@metric"]
      run-groups:
        empty:
          args: []
    dummy-algo-mt:
      docker-tag: ann-benchmarks-sklearn
      module: ann_benchmarks.algorithms.dummy_algo
      constructor: DummyAlgoMt
      base-args: ["@metric"]
      run-groups:
        empty:
          args: []
    ball:
      docker-tag: ann-benchmarks-sklearn
      module: ann_benchmarks.algorithms.balltree
      constructor: BallTree
      base-args: ["@metric"]
      run-groups:
        ball:
          args: &treeargs [[10, 20, 40, 100, 200, 400, 1000]]
    kd:
      docker-tag: ann-benchmarks-sklearn
      module: ann_benchmarks.algorithms.kdtree
      constructor: KDTree
      base-args: ["@metric"]
      run-groups:
        ball:
          args: *treeargs
    BallTree(nmslib):
      docker-tag: ann-benchmarks-nmslib
      module: ann_benchmarks.algorithms.nmslib
      constructor: NmslibNewIndex
      base-args: ["@metric", "vptree"]
      run-groups:
        base:
          # When @args is a dictionary, algorithm instances will be generated
          # by taking the Cartesian product of all of its values.
          args: {"tuneK": 10, "desiredRecall": [0.99, 0.97, 0.95, 0.9, 0.85, 0.8,
                0.7, 0.6, 0.5, 0.4, 0.3, 0.2, 0.1]}
          # This run group produces thirteen algorithm instances:
          # NmslibNewIndex("angular", "vptree", {"tuneK": 10,
          # "desiredRecall": 0.99}), NmslibNewIndex("angular", "vptree",
          # {"tuneK": 10, "desiredRecall": 0.97}), and so on up to
          # NmslibNewIndex("angular", "vptree", {"tuneK": 10, "desiredRecall":
          # 0.1}).
    pynndescent:
      docker-tag: ann-benchmarks-pynndescent
      module: ann_benchmarks.algorithms.pynndescent
      constructor: PyNNDescent
      base-args: ["@metric"]
      run-groups:
        pynndescent:
          args: [[10, 20, 40, 80], [4, 8], [30]]
          query-args: [[1.0, 2.0, 4.0, 8.0]]
  euclidean:
    kgraph:
      docker-tag: ann-benchmarks-kgraph
      module: ann_benchmarks.algorithms.kgraph
      constructor: KGraph
      base-args: ["@metric"]
      run-groups:
        kgraph:
          args: [[1, 2, 3, 4, 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100],
                 {'reverse': -1}, True] # XXX: hard-codes save_index as True
    hnsw(nmslib):
      docker-tag: ann-benchmarks-nmslib
      module: ann_benchmarks.algorithms.nmslib
      constructor: NmslibReuseIndex
      base-args: ["@metric", "hnsw"]
      run-groups:
        M-32:
          # If a run group has an array called @arg-groups instead of one
          # called @args, then every element in that array will be separately
          # expanded before then taking the Cartesian product of all of those
          # expansions.
          #
          # Yes, this is a bit of a hack, but some constructors are weird.
          # (This one used to require that dictionaries be encoded as lists
          # of environment variable-style strings -- ["M=32", "post=2",
          # "efConstruction=400"] -- which didn't work with this at all...)
          arg-groups:
            - {"M": 32, "post": 2, "efConstruction": 400}
            - True # XXX: hard-codes save_index as True
            - {"ef": [20, 30, 40, 50, 60, 70, 80, 90, 100, 120, 140, 160, 200,
              300, 400]}
        M-20:
          arg-groups:
            - {"M": 20, "post": 2, "efConstruction": 400}
            - True
            - {"ef": [2, 5, 10, 15, 20, 30, 40, 50, 70, 80, 120, 200, 400]}
        M-12:
          arg-groups:
            - {"M": 12, "post": 0, "efConstruction": 400}
            - True
            - {"ef": [1, 2, 5, 10, 15, 20, 30, 40, 50, 70, 80, 120]}
        M-4:
          arg-groups:
            - {"M": 4, "post": 0, "efConstruction": 400}
            - True
            - {"ef": [1, 2, 5, 10, 20, 30, 50, 70, 90, 120]}
        M-8:
          arg-groups:
            - {"M": 8, "post": 0, "efConstruction": 400}
            - True
            - {"ef": [1, 2, 5, 10, 20, 30, 50, 70, 90, 120, 160]}
    MP-lsh(lshkit):
      docker-tag: ann-benchmarks-nmslib
      module: ann_benchmarks.algorithms.nmslib
      constructor: NmslibNewIndex
      base-args: ["@metric", "lsh_multiprobe"]
      run-groups:
        base:
          args: {"desiredRecall": [0.99, 0.97, 0.95, 0.9, 0.85, 0.8, 0.7, 0.6, 0.5,
                0.4, 0.3, 0.2, 0.1], "H": 1200001, "T": 10, "L": 50, "tuneK": 10}
    SW-graph(nmslib):
      docker-tag: ann-benchmarks-nmslib
      module: ann_benchmarks.algorithms.nmslib
      constructor: NmslibReuseIndex
      base-args: ["@metric", "sw-graph"]
      run-groups:
        NN-10:
          arg-groups:
            - {"NN": 10}
            - True
            - {"efSearch": [800, 400, 200, 100, 50, 30, 20, 15, 10]}
        NN-5:
          arg-groups:
            - {"NN": 5}
            - True
            - {"efSearch": [30, 25, 20, 15, 10, 5, 4, 3, 2, 1]}
    pynndescent:
      docker-tag: ann-benchmarks-pynndescent
      module: ann_benchmarks.algorithms.pynndescent
      constructor: PyNNDescent
      base-args: ["@metric"]
      run-groups:
        pynndescent:
          args: [[5, 10, 20, 40, 80], [4, 8], [20]]
          query-args: [[1.0, 1.5, 2.0, 4.0, 8.0]]
  angular:
    kgraph:
      docker-tag: ann-benchmarks-kgraph
      module: ann_benchmarks.algorithms.kgraph
      constructor: KGraph
      base-args: ["@metric"]
      run-groups:
        kgraph:
          args: [[1, 2, 3, 4, 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100],
                 {'reverse': -1, "K": 200, "L": 300, "S": 20}, False]
    hnsw(nmslib):
      docker-tag: ann-benchmarks-nmslib
      module: ann_benchmarks.algorithms.nmslib
      constructor: NmslibReuseIndex
      base-args: ["@metric", "hnsw"]
      run-groups:
        M-48:
          arg-groups:
            - {"M": 48, "post": 2, "efConstruction": 800}
            - True # XXX: hard-codes save_index as True
            - {"ef": [50, 70, 90, 120, 160, 200, 400, 600, 700, 800, 1000,
              1400, 1600, 2000]}
        M-32:
          arg-groups:
            - {"M": 32, "post": 2, "efConstruction": 800}
            - True
            - {"ef": [10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 120, 140, 160,
              200, 300, 400, 600, 700, 800, 1000, 1200, 1400, 1600, 2000]}
        M-20:
          arg-groups:
            - {"M": 20, "post": 0, "efConstruction": 800}
            - True
            - {"ef": [2, 5, 10, 15, 20, 30, 40, 50, 70, 80]}
        M-12:
          arg-groups:
            - {"M": 12, "post": 0, "efConstruction": 800}
            - True
            - {"ef": [1, 2, 5, 10, 15, 20, 30, 40, 50, 70, 80]}
    SW-graph(nmslib):
      docker-tag: ann-benchmarks-nmslib
      module: ann_benchmarks.algorithms.nmslib
      constructor: NmslibReuseIndex
      base-args: ["@metric", "sw-graph"]
      run-groups:
        NN-30:
          arg-groups:
            - {"NN": 30}
            - True
            - {"efSearch": [700, 650, 550, 450, 350, 275, 200, 150, 120, 80,
              50, 30]}
        NN-15:
          arg-groups:
            - {"NN": 15}
            - True
            - {"efSearch": [80, 50, 30, 20]}
        NN-3:
          arg-groups:
            - {"NN": 3}
            - True
            - {"efSearch": [120, 80, 60, 40, 20, 10, 8, 4, 2]}
    rpforest:
      docker-tag: ann-benchmarks-rpforest
      module: ann_benchmarks.algorithms.rpforest
      constructor: RPForest
      run-groups:
        base:
          args: [[3, 5, 10, 20, 40, 100, 200, 400],
                [3, 5, 10, 20, 40, 100, 200, 400]]
    pynndescent:
      docker-tag: ann-benchmarks-pynndescent
      module: ann_benchmarks.algorithms.pynndescent
      constructor: PyNNDescent
      base-args: ["@metric"]
      run-groups:
        pynndescent:
          args: [[5, 10, 20, 40, 80, 160], [8], [40]]
          query-args: [[1.0, 2.0, 4.0, 8.0, 16.0, 32.0, 64.0]]

bit:
  hamming:
    kgraph:
      docker-tag: ann-benchmarks-kgraph
      module: ann_benchmarks.algorithms.kgraph
      constructor: KGraph
      base-args: ["@metric"]
      run-groups:
        kgraph:
          args: [[1, 2, 3, 4, 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100],
                 {'reverse': -1, "K": 200, "L": 300, "S": 20}, False]
    hnsw(nmslib):
      docker-tag: ann-benchmarks-nmslib
      module: ann_benchmarks.algorithms.nmslib
      constructor: NmslibReuseIndex
      base-args: ["@metric", "hnsw"]
      run-groups:
        M-48:
          arg-groups:
            - {"M": 48, "post": 2, "efConstruction": 800}
            - True # XXX: hard-codes save_index as True
            - {"ef": [50, 70, 90, 120, 160, 200, 400, 600, 700, 800, 1000,
              1400, 1600, 2000]}
        M-32:
          arg-groups:
            - {"M": 32, "post": 2, "efConstruction": 800}
            - True
            - {"ef": [10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 120, 140, 160,
              200, 300, 400, 600, 700, 800, 1000, 1200, 1400, 1600, 2000]}
        M-20:
          arg-groups:
            - {"M": 20, "post": 0, "efConstruction": 800}
            - True
            - {"ef": [2, 5, 10, 15, 20, 30, 40, 50, 70, 80]}
        M-12:
          arg-groups:
            - {"M": 12, "post": 0, "efConstruction": 800}
            - True
            - {"ef": [1, 2, 5, 10, 15, 20, 30, 40, 50, 70, 80]}
    pynndescent:
      docker-tag: ann-benchmarks-pynndescent
      module: ann_benchmarks.algorithms.pynndescent
      constructor: PyNNDescent
      base-args: ["@metric"]
      run-groups:
        pynndescent:
          args: [[20, 40, 80, 160, 250], [4], [40]]
          query-args: [[1.0, 2.0, 4.0, 8.0, 16.0, 32.0, 64.0]]
int:
  jaccard:
    bf:
      docker-tag: ann-benchmarks-sklearn
      module: ann_benchmarks.algorithms.bruteforce
      constructor: BruteForceBLAS
      base-args: ["@metric"]
      run-groups:
        base:
          args: {}
    datasketch:
      docker-tag: ann-benchmarks-datasketch
      module: ann_benchmarks.algorithms.datasketch
      constructor: DataSketch
      base-args: ["@metric"]
      run-groups:
        base:
          args:  [[16, 32, 64, 128],[5, 10, 20, 40]]